AI Navigate

Andrej Karpathy says humans are now the bottleneck in AI research with easy-to-measure results

THE DECODER / 3/22/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research
Read original →
共有:

Key Points

  • Karpathy argues that humans are now the bottleneck in AI research, with progress increasingly tied to easy-to-measure, scalable improvements.
  • He notes that an autonomous agent was able to optimize his training setup overnight, uncovering improvements he had missed after twenty years of experience.
  • The piece highlights limitations of RLHF and suggests automation could change how researchers experiment and iterate.
  • Overall, the article portrays a shift toward automated optimization and measurement as a driver of AI progress, with broad implications for researchers and product teams.

Former OpenAI researcher Andrej Karpathy criticizes the fact that reinforcement learning from human feedback (RLHF) is only effective to a limited extent when training AI language models.

AI developer Andrej Karpathy let an autonomous agent optimize his training setup overnight, and it found improvements he'd missed despite two decades of experience.

The article Andrej Karpathy says humans are now the bottleneck in AI research with easy-to-measure results appeared first on The Decoder.

💡 Insights using this article

This article is featured in our daily AI news digest — key takeaways and action items at a glance.

📅 3/22DailyView insight →

関連おすすめサービス

※当サイトはアフィリエイト広告を利用しています
🎧

Notta搭載AI議事録イヤホン ZENCHORD1

AI時代の仕事術。Notta搭載で会議の議事録を自動生成するスマートイヤホン。

🎙️

AI搭載ボイスレコーダー Plaud

世界100万人が愛用。AIで文字起こし・要約を自動化するボイスレコーダー。

🖼️

画像高画質化AIツール Aiarty Image Enhancer

AIで画像を高画質化。写真・イラストを簡単にアップスケール。